Apply Previously Estimated Scale Parameters to New Data

When you have scale estimates from a previous fit (via fit_saturated_weight() or fit_saturated_weight_batch()) and want to apply them to a new or subsetted data frame without re-fitting the brms model.

Usage

apply_saturated_weights(
  df,
  scale_table,
  group_vars = c("antigen", "source"),
  concentration_col = "predicted_concentration",
  se_col = "se_concentration",
  pcov_col = "pcov"
)

Arguments

df

Data frame to receive weights.

scale_table

A scale_table from fit_saturated_weight_batch(), or a one-row tibble/data.frame with at least phi and beta1 (or gamma_0 and gamma_1).

group_vars

Character vector: column names to match between df and scale_table for group-specific scale parameters. NULL = apply a single global (phi, beta1) to all rows.

concentration_col

Character: predicted concentration column name.

se_col

Character: SE of concentration column name.

pcov_col

Character: posterior CV column name.

Value

The input data frame with added columns: yi, cv_i, log_cv, cv_source, sigma_i, w_saturated, w_saturated_norm.

Details

This is the Stage 2 function in the recommended two-stage workflow: fit once on the full dataset (all arms, all timepoints), then apply weights to any comparison subset.

See also

fit_saturated_weight_batch() for producing the scale_table, compute_saturated_weights() for the underlying weight computation.

Examples

# \donttest{
data(example_assay)

# Fit on one group first
dat_prn <- example_assay[example_assay$antigen == "prn" &
                         example_assay$feature == "IgG1", ]
dat_prn$cell <- interaction(dat_prn$group_a, dat_prn$group_b, drop = TRUE)
sw <- fit_saturated_weight(dat_prn, cell_col = "cell", pcov_col = "pcov",
                           plate_col = "plate",
                           iter = 1000, warmup = 500, chains = 2, cores = 2)
#> fit_saturated_weight: 506 of 512 observations usable (6 removed); 8 cell levels
#>   cv: OK: sd(log_cv) = 0.487; beta1 identifiable from 506 observations
#>   location: yi ~ 0 + cell + (1 | plate)
#>   scale:    sigma ~ log_cv
#>   fitting brms model (1000 iter, 2 chains)...
#> Warning: Bulk Effective Samples Size (ESS) is too low, indicating posterior means and medians may be unreliable.
#> Running the chains for more iterations may help. See
#> https://mc-stan.org/misc/warnings.html#bulk-ess
#>   sigma fixef rows: sigma_Intercept, cellvaccine_a.timepoint_1, cellvaccine_b.timepoint_1, cellvaccine_a.timepoint_2, cellvaccine_b.timepoint_2, cellvaccine_a.timepoint_3, cellvaccine_b.timepoint_3, cellvaccine_a.timepoint_4, cellvaccine_b.timepoint_4, sigma_log_cv
#>   sigma fixef cols: Estimate, Est.Error, Q2.5, Q97.5
#>   phi = 2.885  [2.03, 4.07]
#>   beta1 = 1.024  [0.829, 1.2]
#>   interpretation: moderate precision weighting
#>   n_eff = 415.1 of 506 (ratio = 0.82)
#>   weight_ratio = 1027  gini = 0.259

# Build a scale_table manually (or use batch$scale_table)
st <- data.frame(antigen = "prn", phi = sw$phi, beta1 = sw$beta1)

# Apply to new data without re-fitting
dat_new <- example_assay[example_assay$antigen == "prn" &
                         example_assay$feature == "IgG1" &
                         example_assay$group_b == "timepoint_3", ]
dat_weighted <- apply_saturated_weights(
  df          = dat_new,
  scale_table = st,
  group_vars  = "antigen",
  pcov_col    = "pcov"
)
summary(dat_weighted$w_saturated_norm)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's 
#>  0.0302  0.8877  1.0943  1.0000  1.2029  1.3718       2 
# }